jax-ml/ml_dtypes

Issues

• Free-threaded wheels on PyPI?

  #228 opened 2 months ago by ngoldbaum
  0

• float8_e4m3fn round 4.25 to 4

  #221 opened 3 months ago by jq
  3

• bfloat16 unsupported by torch.tensor that takes np arrays inputs

  #81 opened a year ago by Ishticode
  5

• [RFC] Change float8_e5m2 dtype descr from <f1 to <V1

  #216 opened 3 months ago by apivovarov
  1

• [Q] How to properly save and load fp8 NumPy arrays?

  #207 opened 3 months ago by apivovarov
  3

• `np.testing.assert_array_equal()` not compatible with bfloat16 when the value is `nan`

  #206 opened 3 months ago by ayaka14732
  1

• [Question] Can we release a new version of ml_dtypes?

  #185 opened 3 months ago by apivovarov
  3

• v0.5.0 release partially failed

  #196 opened 3 months ago by jakevdp
  3

• Support unpacked FP4E2M1

  #164 opened 3 months ago by justinchuby
  5

• Build of ml-dtypes 0.4.0 is failing on Power

  #183 opened 3 months ago by sandeepgupta12
  3

• Add MXFP4 (FP4 E2M1) and MXFP6 (FP6 E3M2/E2M3)

  #116 opened 3 months ago by cloudhan
  13

• Type promotion behavior with Python scalars changed with NumPy 2

  #189 opened 3 months ago by hawkinsp
  0

• np.floor_divide(?, ml_dtypes.???(0.0)) return NaN but np.float16 returns Inf.

  #170 opened 4 months ago by apivovarov
  3

• Updating v0.2.0: Adding new float8 type (fp8e4_m3)

  #165 opened 4 months ago by tapan0007
  2

• Finite precision error calculations always 0 under JIT with bfloat16

  #167 opened 4 months ago by colehaus
  2

• f16 sNaN is casted to f8e5m2 qNaN in Half_To_Float8E5m2 test

  #162 opened 5 months ago by apivovarov
  2

• Update to Numpy 2.0?

  #159 opened 5 months ago by sbrl
  2

• Add complex32 and bcomplex32

  #110 opened a year ago by hawkinsp
  3

• Package Requirements Looks Inconsistent

  #155 opened 6 months ago by daskol
  5

• Support HiF8

  #152 opened 7 months ago by dengguojie
  0

• TBox

  #150 opened 7 months ago by vxst
  1

• Deserialization of float8_e5m2

  #148 opened 8 months ago by wonjeon
  2

• put a description ?

  #149 opened 8 months ago by stonebig
  1

• Support for non-saturating mode for fp8

  #147 opened 8 months ago by wonjeon
  0

• Compilation Error in libtensorflow Pointing to float8.h in ml_dtype

  #144 opened 9 months ago by HongMJ1315
  6

• numpy casting error: Cannot cast array data from dtype(bfloat16) to dtype(bfloat16) according to the rule 'unsafe'

  #145 opened 9 months ago by idanori
  5

• pybind11 version update breaks build

  #83 opened 9 months ago by wsdmatty
  1

• Erroneous issue

  #141 opened 9 months ago by awf
  0

• Feature Request: 8 bit fixed point numbers

  #67 opened 2 years ago by jenkspt
  2

• Adept to NumPy change

  #135 opened 10 months ago by seberg
  1

• bfloat16s (at least) throw error when given a format specifiier

  #134 opened a year ago by colehaus
  0

• Windows builds are failing

  #127 opened a year ago by jakevdp
  1

• Custom floats do not respect `np.set_printoptions`

  #125 opened a year ago by jakevdp
  0

• setup.py has no license information

  #115 opened a year ago by kwijk
  2

• Make a 1.0 release and follow semantic versioning for breaking changes

  #111 opened a year ago by hawkinsp
  0

• Tensorflow build failed after the ml_dtypes new release

  #101 opened a year ago by ShahriyarR
  5

• Test failures on aarch64-darwin

  #106 opened a year ago by samuela
  3

• DType comparisons return Python bools instead of Numpy bools

  #105 opened a year ago by colehaus
  3

• Deprecate creation of dtypes via `np.dtype('typename')`

  #93 opened a year ago by jakevdp
  4

• Add gtest instructions

  #76 opened 2 years ago by awf
  1

• Package vendoring issues

  #51 opened 2 years ago by daskol
  11

• `data` attribute raises a `ValueError`

  #80 opened 2 years ago by joshlk
  0

• Support np.testing.assert_array_max_ulp

  #56 opened 2 years ago by mrahtz
  1

• ml-dtypes does not build from sources on rockylinux 8 with gcc-c++ 8.5

  #69 opened 2 years ago by thebaptiste
  2

• some tests "fail" on osx-arm64, just fyi

  #47 opened 2 years ago by ngam
  7

• Limited bfloat16 precision breaks reduction operations

  #63 opened 2 years ago by mrahtz
  8

• Add Linux aarch64 wheels

  #45 opened 2 years ago by hawkinsp
  0

• Add py.typed marker

  #43 opened 2 years ago by hawkinsp
  0

• bfloat16 & other custom dtypes incompatible with np.save

  #41 opened 2 years ago by jakevdp
  0